QoS Routing in Wireless Mesh Networks with

Cognitive Radios
Roberto Hincapie, Jian Tang, Guoliang Xue and Roberto Bustamante
Abstract In this paper, we study QoS routing in wireless mesh
networks with cognitive radios, which involves route selection,
channel allocation and scheduling. It turns out to be a hard
problem because of the impact of interference and channel
heterogeneity. We formally model it as an optimization problem
and present an Integer Linear Programming (ILP) formulation
to provide optimal solutions. We then present a distributed
routing protocol which can select a route and allocate resources
for a connection request to satisfy its end-to-end bandwidth
requirement. NS2 based simulation results show the performance
given by our protocol is close to that of the optimal solution.
Index Terms Cognitive radio, wireless mesh network, routing,
QoS.

I. I NTRODUCTION
Various multimedia and real-time applications, such as
neighborhood gaming, Video-on-Demand (VoD) and emergency communications, are expected to be provided by Wireless Mesh Networks (WMNs) in the future [2]. In order to
provide those services, end-to-end Quality of Service (QoS)
must be well supported. Bandwidth is a basic QoS parameter
and the focus of this paper. A QoS connection request should
be admitted if there exists a path with required bandwidth.
Otherwise, it should be rejected. Such a bandwidth guaranteed
path in a wired network can be easily obtained by simply
ignoring those links without enough bandwidth and applying
a shortest path algorithm in the residual network. However,
QoS routing in a multihop wireless network is much harder
due to the impact of interference. Communications between
a pair of nodes may consume the bandwidth of neighboring
nodes. The available bandwidth in each node/link is related
to interference caused by its neighbors within the interference
range. Therefore, QoS routing schemes proposed for wired
networks are not applicable for multihop wireless networks.
The emerging cognitive radio technology enables unlicensed
users (a.k.a secondary users) to sense and access the underutilized spectrum bands dynamically as long as the communications among licensed users (a.k.a primary users) in such
spectrum bands are not affected. The cognitive radio is desirable for a WMN in which a large volume of trafc is expected
to be delivered since it is able to utilize spectrums more
efciently, therefore improve network capacity signicantly.
This research is funded in part by NSF grants CNS-0721880 and CNS0721803.
Roberto Hincapie is with the Department of Telecommunications Engineering, Universidad Ponticia Bolivariana, Medellin, Colombia. Email:
roberto.hincapie@upb.edu.co. This work was done when he visited Montana State University. Jian Tang is with the Department of Computer
Science at Montana State University, Bozeman, MT 59717-3880. Email:
tang@cs.montana.edu. Guoliang Xue is with the Department of Computer
Science and Engineering at Arizona State University, Tempe, AZ 85287-8809.
Email: xue@asu.edu. Roberto Bustamante is with the Department of Electric
and Electronic Engineering at Universidad de los Andes, Bogota, Colombia.
Email: rbustama@uniandes.edu.co.

However, it introduces additional complexities to QoS routing.

With cognitive radios, each node can access a set of available
spectrum bands which may spread a wide range of frequencies.
Different spectrum bands (a.k.a channels) can support quite
different transmission ranges and data rates, both of which
have a signicant impact on route selection and end-to-end
bandwidth.
In this paper, we study the QoS routing problem in WMNs
with cognitive radios, which involves not only route selection
but also channel allocation and scheduling. We formally dene
the problem and present an Integer Linear Programming (ILP)
formulation to provide optimal solutions. We then present a
novel distributed routing protocol which can select a route
and allocate channels and timeslots for a connection request
to satisfy its end-to-end bandwidth requirement. The efciency
of the proposed protocol is justied by NS2 based simulation.
To our best knowledge, we are the rst to study QoS routing in
the context of wireless mesh networks with cognitive radios.
II. R ELATED W ORK
Even though cognitive radio came into being a long time
ago, research on networking with cognitive radios is still in its
infancy [3]. In [19], Zheng et al. developed a graph-theoretic
model to characterize the spectrum access problem and devised
a set of heuristics to nd high throughput and fair solutions.
The authors of [16] introduced the concept of a time-spectrum
block and proposed the protocols to allocate such blocks.
In [18], the authors derived optimal and suboptimal distributed
strategies for the secondary users to decide which channels to
sense and access under a framework of Partially Observable
Markov Decision Process (POMDP). In [14], Wang et al.
considered the joint design of dynamic spectrum access and
adaptive power management, and proposed a power-saving
multi-channel MAC protocol. However, the above works
concentrated on the physical and MAC layers and did not
address routing and end-to-end issues, especially end-to-end
QoS provisioning which is the focus of this paper.
QoS routing has been well studied for single-channel
wireless ad hoc networks. QoS routing protocols have been
proposed for wireless ad hoc networks with a TDMA-overCDMA based MAC layer [8], [9], a TDMA-based MAC layer
[13], [20] or a 802.11-based MAC layer [15]. In [12], Tang et
al. proposed an algorithm to compute a low-interference channel assignment as well as a QoS routing algorithm to compute
a route for a connection request with bandwidth requirement
in a 802.11-based multi-channel WMN. However, a cognitive
radio WMN is different from single-channel wireless networks
since different non-overlapping channels can be assigned to
cognitive radios in a common neighborhood to mitigate the
interference. Moreover, a WMN with cognitive radios which

978-1-4244-2324-8/08/$25.00 2008 IEEE.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

can be tuned to a large number of heterogeneous channels is

also quite different from the WMNs studied before [2], [12],
in which several homogeneous channels on a certain spectrum
band are always available to each node.
III. P ROBLEM F ORMULATION
In this section, we will describe our system model rst. Then
we formally dene the problem to be studied and present an
ILP formulation.
We consider a WMN composed of a set of stationary
nodes with locations known, each of which is equipped with
a cognitive radio. The available spectrum is divided into a
set of non-overlapping spectrum bands, which are also called
channels. A cognitive radio can dynamically access a channel
to deliver its packets. Any proposed spectrum sensing scheme
can be used to detect the locally available channels [3].
A TDMA scheme is used in the MAC layer for cognitive
radios, i.e., time domain is divided into timeslots with xed
durations and they are further grouped into frames of T
timeslots each. A timeslot-channel pair (t, h) is dened as
a transmission block, which can be considered as a resource
unit. The capacity of a transmission block c(t, h) = Ch /T ,
where Ch is the capacity of channel h. The resource allocation
problem in the MAC layer is actually to determine how to
allocate available transmission blocks to links subject to the
interference constraints. There also exists a common control
channel on a relatively low frequency, which can support a
large transmission range. Each node is also assumed to have a
low-cost control radio (no need to be a cognitive radio) which
is used for exchanging control messages over the common
control channel.
During a certain period, there will be a set Hv of channels
available to a particular node v. The channel availability may
change over the time due to the impact of primary users
in the same area [3]. We assume every transmitter always
transmits at a xed transmission power level. Hence, there
are a xed transmission range Rh and a xed interference
range Ih (which is typically 2 to 3 times of the transmission
range [11]) for each channel h. Note that the transmission and
interference ranges are both channel-dependent. We model the
network using a communication graph G(V, E), where each
node v V corresponds to a wireless mesh node and there
is a link e E
between nodes u and v if there exists a
channel h Hu Hv and ||u v|| Rh , where ||u v||
represents the Euclidean distance between nodes u and v. We
address wireless interference based on the protocol model [7].
Two links e1 and e2 are said to interfere with each other if
1) e1 is incident to e2 (due to constraints enforced by halfduplexing, unicast communications or collisions) or 2) they
work on the same channel h, and ||T (e1 ) T (e2 )|| Ih or
||T (e1 ) R(e2 )|| Ih or ||R(e1 ) T (e2 )|| Ih , ||R(e1 )
R(e2 )|| Ih . Without confusions, T ()/R() represent both
the transmitter/receiver of the given link, and their locations.
Similar to IEEE 802.11 [1], both the transmitter and receiver
of a link need to be free of interference because we assume
that an ACK packet will be sent back to the transmitter by the
receiver whenever receiving a data packet.

Suppose that channel h Hv . The status of transmission

block (t, h) on node v, S(v, t, h), is said to be transmitting
(T X) if node v transmits on (t, h); receiving (RV ) if v
receives on (t, h); interfered (IR) if v is interfered by another
transmission(s) on (t, h); or free (F R) otherwise. These four
states are disjoint. A transmission block(t, h) is said to be
free for link e = (u, v) if h Hu Hv , ||u v||
Rh , S(u, t, h) = F R, S(v, t, h) = F R and S(x, t, h) =
IR/F R, x (Dhu Dhv )\{u, v}, where Dhu = {x : ||xu||
Ih } and Dhv = {x : ||x v|| Ih }. Note that if the existing
trafc and the corresponding resource allocation are known,
then we can easily identify all the free transmission blocks for
each link based on this denition.
We consider a dynamic trafc model in which connection requests arrive randomly. Each connection request has
a bandwidth requirement B. When a new connection request
arrives, we have to nd a path for routing and reserve resources
(transmission blocks) along the path to satisfy its bandwidth
requirement. We assume that a connection always uses a single
path for packet delivery, i.e., no trafc splitting is allowed.
There are two kinds of potential contentions associated with
a routing path: interow contention and intraow contention
[17]. Interow contentions are the contentions caused by the
interference between a link selected for the new connection
request and another link used for an existing ow. Intraow
contentions are the contentions caused by the interference
between two different links selected for the new connection
request. Interow contentions can be easily resolved if we
always assign aforementioned free transmission blocks (which
can be identied beforehand) to each link on the routing path
we plan to select for the new connection request. Intraow
contentions are much harder to address because we have to
make sure that a free transmission block is not shared by any
two links interfering with each other on the selected path,
which might even be several hops apart.
We now are ready to dene the problem to be studied.
Suppose that we are given a connection request with a source
node s, a destination node d and a bandwidth requirement B.
Denition 1: A feasible transmission schedule for a single
s d path P is a transmission block assignment that allocates
transmission blocks for every link on the path such that
it does not introduce any interow contentions or intraow
contentions, and the bandwidth requirement B is satised on
each link of the path. The path P is said to be an admissible
path if there exists such a feasible transmission schedule.
Denition 2: The Bandwidth Aware Routing (BAR) problem seeks a single sd path along with a feasible transmission
schedule such that the hop count of the path is minimum
among all admissible s d single paths.
Next, we present an ILP formation for the BAR problem.
We dene two decision variables.
xe : xe = 1 if link e is selected for routing; xe = 0,
otherwise.
(t,h)
(t,h)
ye
: ye
= 1 if transmission block (t, h) is allocated
(t,h)
to link e; ye
= 0, otherwise.
We also introduce the following notations.
Fe is the set of free transmission blocks on link e.

978-1-4244-2324-8/08/$25.00 2008 IEEE.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

Evout is the set of outgoing links of node v.

in
Ev is the set of incoming links of node v.
(t,h)
Le
is the set of links which have (t, h) as a free block
and interfere with link e over channel h. Note that e
(t,h)
Le .
ILP : BAR

min
xe
(1)

eE

subject to:

(t,h)Fe

xe

eEvout

xe = 1;

(2)

eEsout

xe = 0,

v V \ {s, d};

(3)

eEvin

ye(t,h) xe ,
c(t, h)

ye(t,h)

xe B,
(t,h)

(t, h) Fe , e E; (4)
e E;

(5)

1,

e E, (t, h) Fe ; (6)

xe {0, 1},
ye(t,h) {0, 1},

(t, h) Fe , e E. (8)

ye

(t,h)
e
Le

e E;

(7)

The objective function (1) is set to minimize the total cost

(hop count) of the path. Constraints (2) and (3) are ow
constraints which guarantee that a single s-d path is chosen.
Constraint (4) establishes a connection between two variables
and ensures that free transmission blocks are only allocated to
the links on the chosen path. Constraint (5) guarantees that the
bandwidth requirement is satised on each link of the chosen
path. Note that c(t, h) is a constant and represents the capacity
of block (t, h). Constraint (6) is the interference constraint
which makes sure there are no intraow contentions among
the chosen links. Note that based on the denition of the free
transmission block, the interow contentions can be prevented
if we always allocate free transmission blocks to the selected
links. Solving the ILP can provide optimal solutions for the
BAR problem, however, it may take a very long time to solve it
for large cases. Hence, we will present an effective distributed
protocol to nd the route and the corresponding transmission
block assignment in the next section.
IV. T HE P ROPOSED ROUTING P ROTOCOL
In order to enable the transmission block allocation, every
node needs to keep track of some necessary information of
all the other nodes in its interference neighborhood, such
as channel availability and transmission block status. In our
protocol, each node will broadcast these information to its
two-hop neighbors over the common control using the control
radio whenever there are any changes. Note that we assume
the common control channel can support a large transmission
range. Hence, every node can obtain these information from all
nodes which can potentially interfere with it. In our protocol,
all the routing control packets are exchanged over the common
channel using the control radios.
Due to the ad hoc feature and highly dynamic properties of
the target network, we propose an on-demand routing protocol.

Similar to Dynamic Source Routing (DSR) [6] and AODV [4],

the route is discovered upon the arrival of a connection request.
Furthermore, the proposed protocol is essentially a source
routing protocol, i.e, the complete or partial routing path
is always included in every control or data packet. In the
DSR, when a connection request arrives at the source node, it
broadcasts an RREQ (Route REQuest) packet to its neighbors.
When an intermediate node receives an RREQ, it must decide
if it should discard or re-broadcast it. Note that for a particular
connection request, it may receive multiple RREQs from
different neighbors, which include different routing paths.
The intermediate node simply re-broadcasts the rst one and
discards all the following. Similarly, the destination node
replies to the source node with an RREP (Route REPly) packet
along the path included in the rst RREQ it receives and
discards all the following.
However, this route discovery procedure cannot well support
end-to-end QoS because it is possible that there does not
exists a feasible transmission schedule along the routing path
included in the rst RREQ arriving at the destination node.
We need to identify more alternative source-destination paths
during the route discovery to increase the chances for the
QoS connection request to be accepted by the network. In
our protocol, an intermediate node v starts a timer whenever
it receives the rst RREQ. For each RREQ received from a
neighbor u before the timer res, node v runs an allocation
algorithm to nd out good feasible transmission blocks for link
e = (u, v). The window size is set to 4ms in the simulation. If
it fails to nd a feasible allocation, it drops the corresponding
RREQ. Otherwise, it attaches itself to the current partial path,
includes it along with the selected blocks in the header, and
re-broadcast it. The details of the proposed transmission block
allocation algorithms will be discussed later.
Similarly, the destination node sets up a window to collect
multiple RREQs for a connection request. It will run the same
allocation algorithm to select transmission blocks for the last
hop. It then chooses an admissible path with minimum hop
count and reply to the source node with an RREP along the
selected path. Accordingly, the transmission blocks selected
for each link along this path will be reserved for transmission.
Moreover, all the other nodes in the neighborhoods of the
nodes in the selected path will be notied that these transmission blocks are going to be used by the new incoming
connection. After the RREP packet arrives at the source node,
data transmission begins. However, if our protocol cannot nd
an admissible path, it will reject the connection request. When
the source or the destination node decides to terminate a
connection, a special control packet RREL (Route RELease)
will be forwarded along the routing path to release the reserved
resources and update the transmission block statuses in the
affected nodes. In addition, similar to DSR, the RERR (Route
ERRor) based failure recovery scheme will be used to notify
the source node and initiate a new route discovery after
node failures and/or link failures (usually caused by channel
availability change).
Next, we present two heuristic algorithms to allocate transmission blocks for the link e = (u, v). We use B  to denote
the remaining bandwidth requirement. Initially, B  = B.

978-1-4244-2324-8/08/$25.00 2008 IEEE.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

(t,h)

e
(t,h)

impact ratio as Iint (t, h) = max{0, 1 e Le (t,h)

}, to
|Le
|
indicate the impact of selecting block (t, h) for link e on its
interference neighborhood. A relatively small value indicates
that block (t, h) is one of few free transmission blocks in the
neighborhood, i.e., if block (t, h) is chosen for link e, those
links will not be able to accept more ows, which is obviously
not preferred. On the contrary, a relatively large value indicates
that selecting block (t, h) for link e would not seriously
affect the availability of free blocks on the neighboring links.
We now dene our weight function for the free transmission
blocks as follows.
w(t, h) = Icap (t, h) + Iint (t, h)

of 600m. In all simulation scenarios, we randomly placed 10

primary users in the region. Each of them randomly chose a
channel to use, which was then considered to be unavailable
for all the nodes within the corresponding interference range.
In all simulation runs, we injected 500 connection requests
with random source nodes and destination nodes. The mean
inter-arrival time is set to 10. The acceptance ratio is used as
the performance metric, which is dened as the ratio between
the number of accepted connection requests and the total
number of connection requests (i.e., 500).
In the rst scenario, we randomly placed 10 nodes in a
1000m 1000m region. The number of channels (H) was set
to 12 and the number of timeslots in a frame (T ) was set to
10. We divided those channels into three groups with equal
sizes. The corresponding block capacities were normalized to
3, 1 and 0.5 respectively, and the transmission ranges were
set to 200m, 250m, and 500m respectively. The connection
duration was set to a random integer uniformly distributed in
[1, Dmax ], where Dmax = 250 in this scenario. The bandwidth
requirement of a connection request was set to a random
number in [1, Bmax ], where Bmax was changed from 1.25
to 12.5. In scenarios 2 4, we tested larger cases by setting
the region size to 2000m 2000m, H = 24 and T = 20,
which give 480 different transmission blocks. Block capacity
and transmission range settings were the same as those in
scenario 1. In scenario 2, we xed the number of nodes n = 40
and Dmax = 250. Bandwidth requirements were generated
in the same way as in scenario 1. In scenario 3, n = 40,
Bmax = 5, and Dmax was changed from 150 to 550. Finally,
in scenario 4, Dmax = 250, Bmax = 5 and we changed n
from 10 to 90. All the results are presented in Figs. 14. In the
gures, Cap and Cap Interf stand for our capacity based
and capacity-interference based algorithms respectively. The
optimal solutions are provided by solving the ILP presented
in Section III using CPLEX 9.0 [5].
90

Acceptance ratio (%)

Capacity based algorithm: We sort all the free transmission

blocks in the descending order of their capacities. If the
capacity of the unselected free block with the largest capacity
(the rst one in the sorted list) c(tmax , hmax ) B  , then it is
selected. Otherwise, the transmission block with the minimum
capacity among all the free unselected blocks whose capacities
are no less than B  is selected. Then the selected blocks
are removed from the list and B  is updated. This procedure
continues until the bandwidth requirement is satised.
The capacity based algorithm is a simple greedy algorithm whose design philosophy is to select minimum number
of free transmission blocks to satisfy the given bandwidth
requirement. However, the algorithm does not address the
impact of interference. Therefore, we propose another heuristic
algorithm with consideration for both capacity and interference
in the following.
Capacity-interference based algorithm: The basic idea is to
select high capacity transmission blocks, but to avoid using
those transmission blocks which are the only free blocks on
the other links within the interference range. We dene the satisfaction ratio as Icap (t, h) = max{1, c(t,h)
B }, which indicates
how much bandwidth requirement is satised if transmission
(t,h)
block (t, h) is chosen. For a link e Le , where e = (u, v)
(t,h)
and Le
denotes the set of links which have (t, h) as a
free block and interfere with e over channel h, we compute
the total capacity of its free transmission blocks as Ce
=

c(t,h)


(t,h)Fe
c(t, h). We also compute e (t, h) = Ce
, which
indicates the impact of selecting block (t, h) for link e on a
link interfering with e. Furthermore, we dene
the interference


(9)

In the equation, and are tunable parameters which are

used to achieve a selection tradeoff between capacity and the
impact of interference. Moreover, + = 1. We sort free
transmission blocks in the descending order of their weight
values and select the rst k transmission blocks whose total
capacity is no smaller than B to serve the new incoming
connection request on link e = (u, v).
V. S IMULATION R ESULTS
We evaluated the performance of the proposed protocol
via simulation using Network Simulator 2 (NS2) [10]. In the
simulation, the interference ranges were always set to 2 times
of the corresponding transmission ranges [11] for all channels.
The common control channel can support a transmission range

80
70

Cap
CapInterf, =0.5, =0.5
CapInterf, =0.0, =1
Optimal

60
50
40
30
1.5 2.5
5
7.5
10
12.5
The maximum bandwidth requirement (Bmax)

Fig. 1.

Scenario 1: The proposed protocol VS. the optimal solution

We make the following observations:

1) From Fig. 1, we can see that the performance given by
our protocol is very close to that of the optimal solution. Our
capacity-interference based algorithm performs slightly better
than our capacity based algorithm. By setting = 0.5 and
= 0.5, it achieves an average acceptance ratio of 58.09%,

978-1-4244-2324-8/08/$25.00 2008 IEEE.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

Acceptance ratio (%)

100
90

occupation) for the accepted connections. Therefore, it is more

likely that the system rejects more future connection requests.
However, we observed from Fig. 4 that the acceptance ratio
does not necessarily increase with the network size. On the
positive side, with more nodes in the network, there will
exist more alternative paths for a connection request. On the
negative side, more nodes lead to a denser network, therefore
stronger interference.

Cap
CapInterf, =0.5, =0.5
CapInterf, =0.0, =1

80
70
60
50
1.5 2.5
5
7.5
10
12.5
The maximum bandwidth requirement (Bmax)

Fig. 2.

Scenario 2: The impact of bandwidth requirement

100

Acceptance ratio (%)

95
90

Cap
CapInterf, =0.5, =0.5
CapInterf, =0.0, =1

R EFERENCES

85
80
75
70
65
60
150
250
350
450
550
The maximum connection duration (Dmax)

Fig. 3.

Scenario 3: The impact of connection duration

Acceptance ratio (%)

100
Cap
CapInterf, =0.5, =0.5
CapInterf, =0.0, =1
95

90

85
10

Fig. 4.

30
50
70
The number of nodes

VI. C ONCLUSIONS
In this paper, we studied QoS routing in wireless mesh
networks with cognitive radios. We formally modeled it as
an optimization problem. We presented an Integer Linear Programming (ILP) formulation as well as a distributed routing
protocol to solve it. NS2 based simulation results have showed
the performance given by our protocol is close to that of the
optimal solution.

90

Scenario 4: The impact of network size

compared to 59.66% given by solving the ILP. The difference

is only 1.57%.
2) The capacity-interference based algorithm outperforms
the capacity based algorithm by more than 3% on average.
Interestingly, setting = 0.0 and = 1.0 gives slightly
better performance than setting = 0.5 and = 0.5, which
indicates that the interference plays a key role.
3) As expected, we nd out that the acceptance ratio
decreases with the increase of bandwidth requirement and
connection duration. Higher bandwidth requirements (longer
durations) lead to more resource consumption (longer resource

[1] IEEE 802.11 Working Group, Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) Specications, IEEE Standard, 1997.
[2] I. F. Akyildiz, X. Wang and W. Wang, Wireless mesh networks: a survey,
Journal of Computer Networks, Vol. 47, No. 4, 2005, pp. 445487.
[3] I. F. Akyildiz, W-Y Lee, M. C. Vuran and S. Mohanty, NeXt generation/dynamic spectrum access/cognitive radio wireless networks: a
survey, Journal of Computer Networks, Vol. 50, No. 13, 2007, pp. 2127
2159.
[4] C. Perkins, E. Belding-Royer and S. Das, Ad hoc On-Demand Distance
Vector (AODV) Routing, IETF RFC 3561, 2003.
[5] ILOG Software Inc., CPLEX 9.0, http://www.ilog.com
[6] D. Johnson, Y. Hu and D. Maltz The Dynamic Source Routing Protocol
(DSR) for Mobile Ad Hoc Networks for IPv4, IETF RFC 4728, 2007.
[7] P. Gupta and P. R. Kumar, The capacity of wireless networks, IEEE
Transactions on Information Theory, Vol. 46, No. 2, 2000, pp. 388
404.
[8] C. R. Lin and J. S. Liu, QoS routing in ad hoc wireless networks, IEEE
Journal on Selected Areas in Communications, Vol. 17, No. 8, 1999,
pp. 1426-1438.
[9] H. Liu, X. Jia, D. Li and C. H. Lee, Bandwidth guaranteed call
admission in TDMA/CDMA ad hoc wireless networks, Journal of Ad
Hoc Networks, Vol. 3, No. 6, 2005, pp. 689701.
[10] Network Simulator 2 (NS2), http://www.isi.edu/nsnam/ns/
[11] A. Raniwala and T. Chiueh, Architecture and algorithms for an IEEE
802.11-based multi-channel wireless mesh network, Proceedings of
IEEE Infocom2005, pp. 22232234.
[12] J. Tang, G. Xue and W. Zhang, Interference-aware topology control and
QoS routing in multi-channel wireless mesh networks, Proceedings of
ACM MobiHoc2005, pp. 6877.
[13] J. Tang, G. Xue and C. Chandler, Interference-aware routing and bandwidth allocation for QoS provisioning in multihop wireless networks,
Journal of Wireless Communications and Mobile Computing (WCMC),
Vol. 5, No. 8, 2005, pp. 933944.
[14] J. Wang, Y. Fang and D. Wu, A power-saving multi-radio multi-channel
MAC protocol for wireless local area networks, Proceedings of IEEE
Infocom2006.
[15] Q. Xue and A. Ganz, Ad hoc QoS on-demand routing (AQOR) in mobile
ad hoc networks, Journal of Parallel Distributed Computing, Vol. 63,
No. 2, 2003, pp. 154165.
[16] Y. Yuan, P. Bahl, R. Chandra, T. Moscibroda and Y. Wu, Allocating
dynamic time-spectrum blocks in cognitive radio networks, Proceedings
of ACM MobiHoc2007, pp. 130-139.
[17] H. Zhai, J. Wang and Y. Fang, Distributed packet scheduling for
multihop ows in ad hoc networks, Proceedings of IEEE WCNC2004,
pp. 10811086.
[18] Q. Zhao, L. Tong and A. Swami, Decentralized cognitive MAC for
dynamic spectrum access, Proceedings of IEEE DySPAN2005, pp. 224232.
[19] H. Zheng and C. Peng, Collaboration and fairness in opportunistic
spectrum access, Proceedings of IEEE ICC2005, pp. 3132-3136.
[20] C. Zhu and S. Corson, QoS routing for mobile ad hoc networks,
Proceedings of IEEE Infocom2002, pp. 958967.

978-1-4244-2324-8/08/$25.00 2008 IEEE.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.